CS , Fall 1998 Final Exam

Transcription

1 Full Name: Andrew ID: CS , Fall 1998 Final Exam December 11, 1998 Instructions: æ Make sure that your exam is not missing any sheets, then write your full name and Andrew ID on the front. æ Write your answers in the space provided below the problem. If you make a mess, clearly indicate your final answer. æ The exam has a maximum score of 100 points. æ The problems are of varying difficulty. The point value indicated for each problem is proportional to its difficulty. Pile up the easy points quickly and then come back to the harder problems. æ Each problem tests a skill from a lab or homework assignment. The problems are listed in course order. æ This exam is OPEN BOOK. You may use any books or notes you like. Good luck! 1: /05 2: /10 3: /18 4: /10 5: /14 6: /10 7: /16 8: /17 TOTAL: /100 Page 1 of 13

2 Problem 1. (5 points): Write a bit mask generator. Assume that 0 ç low ç high ç 31. Examples: make_mask(12,9) should return 0x000000E0, while make_mask(31,31) should return 0x int make_mask(unsigned int high, unsigned int low) { int mask; return mask; Page 2 of 13

3 Problem 2. (10 points): Consider the following Alpha assembly code for a procedure foo(): foo: 0x0: 47ff0413 bis zero, zero, a3 0x4: ee20000c ble a1, 0x38 0x8: a ldq t0, 0(a0) 0xc: a2 cmpeq a2, t0, t1 0x10: f bne t1, 0x38 0x14: s8addq a3, a0, v0 0x18: b7e00000 stq zero, 0(v0) 0x1c: addq a3, 0x1, a3 0x20: addq v0, 0x8, v0 0x24: a3 cmplt a3, a1, t2 0x28: e beq t2, 0x38 0x2c: a ldq t3, 0(v0) 0x30: a5 cmpeq a2, t3, t4 0x34: e4bffff8 beq t4, 0x18 0x38: bis a3, a3, v0 0x3c: 6bfa8001 ret zero, (ra), 1 Based on the assembly code above, fill in the blanks below in its corresponding C source code. (Note: you may only use symbolic variables a, len, val, andi, from the source code in your expressions below do not use register names.) long int foo(long int *a, long int len, long int val) { long int i; for (i = ; ; i = ) { a[i] = 0; return i; Page 3 of 13

4 Problem 3. (18 points): Dr. Evil has returned! He has placed a binary bomb in this exam! Once again, Dr. Evil has made the disasterous mistake of leaving behind some of his source code. Can you save all of mankind (or at least your grade on this question), and tell us what this bomb does? Turn the page to find out how... The C source code Dr. Evil forgot to erase: /* bomb.c: Use new computer technology to blow up exams! -- Dr. Evil */ #include <stdio.h> #include <stdlib.h> extern long secret_unsolvable_puzzle_fn(long input); void explode_bomb() { printf("you fail! exit(8); Mwhahahahaha!!!\n"); int main(int argc, char *argv[]) { if(argc!= 2) { printf("usage: %s <magic password>\n", argv[0]); explode_bomb(); if(secret_unsolvable_puzzle_fn(atol(argv[1]))!= 1) { explode_bomb(); printf("curses, foiled again! Good work!\n"); return 0; The Alpha dissasembly for the stuff Dr. Evil did erase: secret_unsolvable_puzzle_fn: 0x : addq a0, a0, t0 0x : lda t1, 15213(zero) 0x : cmpult t0, t1, t2 0x c: beq t2, 0x c 0x : addq t0, a0, t0 0x : cmpult t0, t1, t3 0x : bne t3, 0x x c: lda v0, (t0) 0x : cmpeq zero, v0, v0 0x : ret zero, (ra), 1 Page 4 of 13

5 (a) Does the bomb use floating point arithmetic? (Yes/No) (b) Does the function secret_unsolvable_puzzle_fn() contain any recursion? (Yes/No) (c) You think that secret_unsolvable_puzzle_fn() is too slow, and you want to replace it by a lookup table which exhaustively lists all possible inputs and maps them to all possible outputs. How many entries will your lookup table have? What values can secret_unsolvable_puzzle_fn() return? entries in table = possible return values = (d) Just to see what happens, you try running the program./bomb 0. What happens? Does the bomb explode? (e) For each of the following input values, circle whether it defuses or explodes the bomb: input = 2 defuses explodes input = 11 defuses explodes input = 1381 defuses explodes input = 5071 defuses explodes Page 5 of 13

6 Problem 4. (10 points): You have been assigned the task of writing a C function to compute a floating point representation of 2 x. You realize that the best way to do this is to directly construct the IEEE single precision representation of the result. When x is too small, your routine will return 0:0. When x is too large, it will return +1. Fill in the blank portions of the following code to compute the correct result: /* Compute 2**x */ float fpwr2(int x) { union { unsigned u; float f; result; unsigned exp, sig; /* Result exponent and significand */ if (x < ) { /* Too small. Return 0.0 */ exp = ; sig = ; else if (x < ) { /* Denormalized result */ exp = ; sig = ; else if (x < ) { /* Normalized result. */ exp = ; sig = ; else { /* Too big. Return +oo */ exp = ; sig = ; result.u = exp << 23 sig; return result.f; Page 6 of 13

7 Problem 5. (14 points): Welcome to the C-Memory-Layout Question. The context for all questions is the Alpha, as usual. struct s1 { char c; short s; double d; int i[4]; ; struct s2 { int a[2][3]; struct s1 b; long c[2][2]; ; struct s2 thing; a) Fill in the following table to describe the memory footprint of thing. Do not include wasted space in length. To help you, we ve filled in two of the blanks (well, ok - only one is useful). Field Start Length a 0 b.c b.s b.d b.i c 56 b) In an effort to create job security for himself, a deranged programmer at your company had optimized his code by referencing fields of thing via out-of-range elements of thing.a and thing.c. Needless to say, as soon as this was discovered, he was fired. Now, you must figure out what the programmer was actually referencing. First, figure out the offset of the access from the beginning of the array accessed. For example, the offset of thing.b.i[1] is 4. Using the offset and your knowledge of the structure s layout, you can figure out how the programmer should have referred to those memory locations. Write one or more expressions indicating what fields were being referenced. Array indices you give should be within the bounds as given in the declaration. It may help you to draw a picture on scratch paper. Store-to Relative Offset Fields/Elements Overwritten thing.a[3][2] thing.c[-1][-3] thing.b.i[6] Page 7 of 13

8 Translation Lookaside Buffers Translation Lookaside Buffers (TLBs) are often managed by the operating system virtual memory manager using special instructions to update TLB entries. If there are bugs in this software, chaotic behavior can occur. Consider the task of writing a consistency checker that would scan through the TLB and detect anomalous entries. Assume a virtually-addressed memory system with the following properties: æ Virtual addresses are 14 bits wide. æ Physical addresses are 12 bits wide. æ Thepagesizeis64bytes. æ The TLB is 8-way set associative with 16 total entries. æ There should be no aliasing in the address mapping, i.e., under no condition should two virtual addresses map to the same physical address. Observe that for this virtual address format, the upper 8 bits denote the virtual page number (VPN), while the lower 6 bits denote the virtual page offset (VPO). For indexing into the TLB, the low order bit of the VPN indexes the set (TLBI), and the upper 7 bits form the tag (TLBT), as diagrammed below: VPN VPO TLBT TLBI The physical address is split into two fields: the upper 6 bits denote the physical page number (PPN), while the lower 6 bits denote the physical page offset (PPO). The page table (in hexadecimal) for the first 16 pages is as follows: VPN PPN Valid VPN PPN Valid B A B 1A A 0 0C D 2D E F 0D 1 Page 8 of 13

9 Problem 6. (10 points): In this problem, you are given a series of TLB entries for set index 1. For each entry, indicate the virtual page number (VPN) represented (in hexadecimal), whether the entry is valid (i.e, there is nothing wrong with its format or with respect to the portion of the page table given), and a brief explanation. For valid entries, your explanation should describe what the combination of TLB and page table indicates about this page. For invalid entries, you should describe why they are invalid. A. Tag PPN Valid VPN: OK?: Explanation: B. Tag PPN Valid 06 2D 1 VPN: OK?: Explanation: C. Tag PPN Valid VPN: OK?: Explanation: D. Tag PPN Valid 1F 33 1 VPN: OK?: Explanation: E. Tag PPN Valid F VPN: OK?: Explanation: Page 9 of 13

10 Caches In the next problem you will determine the number of read and write misses for a small transpose routine, given data caches of sizes 64 and 256 bytes. Throughout this problem you should assume the following: æ Like Lab L3, the data cache is direct mapped, write through, write allocate, and the block size is 32 bytes. æ The src array starts at address 0 and the dst array starts at address 128. æ Accesses to the src and dst arrays are the only sources of read and write misses, respectively. æ Unlike Lab L3, the arrays are 2d arrays of longs, where sizeof(long) = 8 bytes. Page 10 of 13

11 Problem 7. (16 points): typedef long array[4][4]; void foo(array dst, array src) { int i, j; for (i = 0; i < 4; i++) { for (j = 0; j < 4; j++) { dst[j][i] = src[i][j]; A. For a cache with a total size of 64 data bytes and for each i and j, indicate whether each access to dstëiëëjë and srcëiëëjë is a hit (h) or a miss (m). For example, reading srcë0ëë0ë is a miss and writing dstë0ëë0ë is also a miss. dst array m src array m B. Repeat part A for a cache with a total size of 256 data bytes. dst array src array Page 11 of 13

12 Problem 8. (Of course, 17 points): The designers of the cookienet protocol just can t stop: after their proprietary protocol was reverse engineered, they designed CookieNet In this problem, you ll reverse engineer this protocol. They ve advertised that they ve improved encryption by varying the key on every packet, and increased the number of simultaneous cookie messages possible. You start by using the manufacturer s program, pfw2k to write the following message to the network, specifying port 0x : the quick brown fox 1 the quick brown fox 2222 the quick brown fox Here s a dump of the packets that were sent to the network, both in hex and the character representation, without the Ethernet header. packet 1: B F 77 6E F A...4Vx...the quick brown fox 1. packet 2: A 64 6C F A...4Vx...uif rvjdl cspxo gpy packet 3: D A B 65 6D A A...4Vx...vjg swkem dtqyp hqz a) Name the fields present in the CookieNet 2000 packet header. Does any field serve more than one purpose? Page 12 of 13

13 b) Fill in the C structure below with the format of the packet. If you need a placeholder, clearly label it as such. struct packet { ; unsigned char payload[1]; /* actual payload may be longer */ c) Write, in hex, the header for a packet that contains 0x12C (300) characters and has sequence number 0x10000 (65536), on port number 0x19 (25). You may not need all the spaces. Put one byte per box. Indicate where the start of the payload would be Page 13 of 13